Economists derive optimum solutions to economic problems by assuming unlimited human capacity to maximize (or minimize). This assumption is commonly known as the rationality postulate. Although this approach does not illuminate how human agents actually make economic decisions, the solutions are defended on the basis of empirical testability. In other words, if actual human behavior can be reliably predicted to conform to the optimum solutions, then economic agents can be inferred as if they are able to maximize.

But economic behavior may be predictable exactly because of limited human capacity to maximize when faced with uncertainty. This uncertainty results from human failure to properly distinguish environmental situations and to choose the right actions even if the situations are correctly identified. Regularities of human behavior can be understood as "behavioral rules" to restrict the flexibility to choose potential actions. These mechanisms simplify behavior to less-complex patterns, which are easier for an observer to predict. In the special case of no uncertainty, the behavior of perfectly informed, fully optimizing agents responding with complete flexibility to every perturbation in their environment would not produce easily recognizable patterns, but rather extremely unpredictable behavior.

The criterion to restrict behavioral flexibility is mandated by the need to reduce mistakes in selecting wrong actions under uncertainty. It boils down to the requirement that the expected gain from selecting action A when the situation is right for A exceeds the expected loss from selecting action A when the situation is wrong for A. This criterion can be called the reliability condition. This condition explains why an agent's behavioral repertoire must be limited to actions which are adapted only to relatively likely situations. Thus, a general characteristic of such a repertoire is that it excludes actions which in fact enhance performance under some rare conditions. Therefore, predictable behavior is not an "as if" simulation to optimizing, but rather evolves only to the extent that agents are unable to maximize because of uncertainty. The greater the uncertainty, the more limited the repertoire and the more predictable the behavior.

The reliability condition also explains the existence of market organizations and other social institutions that regularize complex interactions between agents in recurrent situations. On the other hand, neoclassical optimization models that assume away the gap between competence and difficulty largely operate in an institutional vacuum as if the institutional context does not affect the optimum solutions.

The reliability condition is also consistent with many observed behavior which have been treated as "anomalies" by mainstream optimization theory. For example, the inability of businessmen to ignore sunk costs, optimize on the basis of marginal cost, and the stickiness of behavioral response to infrequent shocks.

Constrained behavioral flexibility of individual decision units, however, does not rule out more sophisticated behavior in complex organizations. In fact, it is the very predictability of lower level structures that allow more complex behavioral repertoire to emerge in higher-level structures. In turn, reliability of lower-level structures is sustained because they are required to consider only local information.